Regression Model for Civil Aero-engine Gas Path Parameter Deviations Based on Res-BP Neural Network

Abstract

The gas path parameter deviations as crucial parameters can assist each airline to realize the performance state trend analysis, life prediction and fault diagnosis of aero-engine. However, the calculation of gas path parameter deviations is complicated and the calculation models are also mastered by the original equipment manufacturer (OEM), which makes it burdensome for airlines to independently analyze the gas path performance of the aeroengine. At present, airlines have accumulated a large number of samples of gas path parameter deviations, which makes it possible to establish a regression model between gas path parameters and its deviations by data-driven method. In order to enhance the analysis capability of airline in gas path performance, we apply the residual learning blocks to the back propagation (BP) neural network based on the learning mechanism of the residual networks (ResNets). According to the solution characteristics of gas path parameter deviations, the regression models for the gas path parameter deviations are established based on Res-BP neural network. The screening for nonlinear independent variables of regression model is carried out by mean impact value (MIV) method, and then the input and output of Res-BP neural network can be determined. After the regression model training, the test set is tested by the proposed regression model. By comparing with BP neural network regression model and traditional regression model, the proposed regression model manifests higher prediction accuracy and generalization performance on the three key gas path parameter deviations, which is of great guiding significance for the aero-engine condition monitoring.